Trump's AI Order Asks Permission, and That's Exactly What Industry Wanted

Musk, Zuckerberg, and David Sacks killed the 90-day review window overnight. What's left is a voluntary framework with no teeth, no privacy protections, and no restrictions on how the government uses AI to surveil Americans.

State of Surveillance Editorial Team

The White House building in Washington DC viewed from the South Lawn

TL;DR: On June 2, 2026, President Trump signed an executive order titled “Promoting Advanced Artificial Intelligence Innovation and Security.” It asks AI companies to voluntarily let the government test their most powerful models for up to 30 days before public release. That’s it. No mandatory compliance. No enforcement mechanism. No privacy protections. No restrictions on how agencies use AI for surveillance. The original draft gave the government 90 days, until Elon Musk, Mark Zuckerberg, and former AI czar David Sacks called Trump the night before a planned signing ceremony and killed it. What survived is the weakest possible version of AI oversight: an order that asks nicely and explicitly bans itself from ever becoming a real rule.

Three Phone Calls, One Dead Policy

Trump was supposed to sign this order on May 21, 2026. A White House ceremony was scheduled. Press had been briefed. The original draft gave the government up to 90 days to test “covered frontier models” before companies could release them: a timeline that national security officials had pushed for to actually assess whether an AI system could be weaponized against critical infrastructure [1].

Then, between the night of May 20 and the morning of May 21, three people called the president: Elon Musk, Mark Zuckerberg, and David Sacks, the venture capitalist who had just stepped down as White House AI czar in late March [2].

Their argument was simple: even a voluntary 90-day window would slow American AI companies relative to Chinese competitors. Trump canceled the ceremony hours before it was set to begin, telling reporters: “I didn’t like certain aspects of it” [2].

What he meant was: three billionaires didn’t like certain aspects of it.

What Actually Survived

The order Trump signed twelve days later, on June 2, is a different animal. It was signed privately: no ceremony, no livestream, no photo op with Silicon Valley CEOs [1]. Here’s what’s in it:

  • 30-day voluntary review: AI companies are asked to let the government test their most powerful models for up to 30 days before public release. Industry had pushed for two weeks. National security officials wanted 90 days. The 30-day window was the compromise [3].
  • Cybersecurity benchmarks: Federal agencies have 60 days to develop a classified process for identifying “covered frontier models” and assessing their cyber capabilities. The NSA runs the classified designation process [4].
  • AI cybersecurity clearinghouse: A Treasury-led coordination body linking government, AI firms, and infrastructure operators to find and patch software vulnerabilities [5].
  • DOJ enforcement: The Attorney General must prioritize prosecution of anyone using AI to “illegally access or damage computer systems, steal data, or facilitate other criminal activity” [5].

And here’s the kicker, the sentence that tells you everything about who wrote this order:

“Nothing shall be construed to authorize creation of any mandatory governmental licensing, pre-clearance, or permitting requirement” for AI model development [5].

The order doesn’t just fail to create mandatory oversight. It explicitly bans itself from ever becoming mandatory oversight. It’s a rule that says it’s not allowed to be a rule.

Sacks Wrote the Escape Hatch

David Sacks left his official role as White House AI czar in late March 2026. He now co-chairs the President’s Council of Advisors on Science and Technology. But he kept writing the policy even after leaving office, securing the shorter review window, the voluntary framework, and the anti-mandatory-licensing language alongside National Economic Council deputy director Ryan Baasch [3].

After the signing, Sacks called the order “a game changer” [3]. He also went on record insisting it was “not an FDA for AI” (a telling denial, since nobody had accused it of being one) [6].

Here’s the split that matters: OpenAI actually supported the original 90-day order through its chief lobbyist, putting it in direct opposition to Musk and Zuckerberg [2]. That’s a rare case where the surveillance state’s usual suspects don’t align neatly. OpenAI wanted tougher rules, possibly because mandatory testing creates a moat that benefits incumbents over startups. The motives were different. The outcome was the same: industry won.

What’s Not in the Order: Everything That Matters

From a surveillance and privacy standpoint, this order is a blank check. Here’s what it doesn’t include:

  • No restrictions on government AI surveillance. Nothing in the order limits how federal agencies (ICE, the FBI, DHS, the NSA) use AI to monitor Americans. The focus is entirely on testing models for cybersecurity risks to infrastructure. How the government itself deploys AI against its own citizens? Not addressed [5].
  • No privacy protections. The White House fact sheet doesn’t mention privacy once. Vice President Vance referenced “Americans’ data and privacy” in passing, but the actual order contains no privacy provisions [5].
  • No bias requirements. Biden’s revoked October 2023 AI order (EO 14110) included mandates to prevent biased AI in government decision-making. Trump revoked that order on his first day in office, January 20, 2025. This new order doesn’t replace any of those protections [7].
  • No transparency requirements. Companies aren’t required to disclose what data they trained on, how their models make decisions, or who has access to the outputs.

Senator Mark Warner raised the alarm that the administration could “politicize” the voluntary testing regime to pressure companies on unrelated policy objectives [3]. Given that Trump already banned the entire federal government from using Anthropic’s products after the company refused to waive its restrictions on mass surveillance and autonomous weapons, that concern is well-founded [8].

The Anthropic Problem

The order says it relies on “voluntary collaboration from America’s leading AI companies, like Anthropic, OpenAI, and Google” [1]. But the administration is currently at war with one of those companies.

In February 2026, the Pentagon moved to designate Anthropic a national security “supply chain risk” after the company refused to let its AI systems be used for lethal autonomous weapons or mass domestic surveillance of Americans. Defense Secretary Hegseth declared the company’s ethical guardrails a risk to national security. Trump ordered every federal agency to “immediately cease all use of Anthropic’s technology” [8].

So the same administration that punished a company for refusing to participate in mass surveillance is now asking that company to voluntarily submit its models for government testing. The contradiction writes itself.

This is what the order really is: a framework where companies that cooperate with the surveillance state get access and favor, and companies that draw ethical lines get blacklisted. It’s not oversight. It’s selection pressure.

Biden’s Order Was Stronger. Trump Revoked It.

For context: Biden’s October 2023 AI executive order (EO 14110) was the most ambitious federal AI policy the U.S. had ever attempted. It required AI developers to share safety test results with the government before launch. It mandated watermarking of AI-generated content. It directed agencies to address bias in AI systems used in housing, criminal justice, and healthcare. It created the AI Safety Institute at NIST [7].

Trump revoked it on January 20, 2025 (his first day in office) calling it a barrier to innovation [7].

What replaced it eighteen months later asks companies to voluntarily do less than Biden required them to do. The CFR assessment put it bluntly: the order is “best understood as an attempt to engineer a cybersecurity window of opportunity” but represents only a “first step” that needs a “more comprehensive approach” [4]. That’s diplomatic language for: this doesn’t do much.

The Real Problem: AI Is Already Watching You

While the White House debated how many days to give companies before a voluntary test that nobody has to participate in, federal agencies are already using AI in ways that would have seemed dystopian a decade ago:

  • ICE uses AI-powered surveillance to track immigrants across cities using license plate readers, social media monitoring, and cell phone location data.
  • The FBI has used facial recognition against protest photos without warrants.
  • The Pentagon’s budget includes $9 billion for AI and advanced computing, with a significant share going to intelligence agencies [9].
  • The Trump administration is reportedly building a centralized AI database to consolidate information on every American across federal agencies [10].

None of this is touched by the June 2 executive order. The order looks outward (at what companies build) and asks politely. It does not look inward at what the government does with AI it already has. That silence is the policy.

What You Can Do

  • Don’t mistake this for progress. A voluntary framework with no enforcement and no privacy provisions is not AI regulation. It’s PR.
  • Watch the SECURE Data Act: The House held a hearing today (June 3) on H.R. 8413, which would create a federal privacy framework. Whether it passes and what it includes matters far more than this executive order.
  • Push for state-level AI regulation: States are moving faster than Congress on AI privacy: surveillance pricing bans, facial recognition restrictions, biometric data protections. Your state legislature is where the real action is.
  • Support organizations fighting AI surveillance: The EFF, EPIC, and ACLU are the ones doing the actual work of challenging government AI surveillance in court. They need funding.
  • Track which companies participate: The order is voluntary. Watch which companies submit models and which don’t, and what they get in return from the administration.

Sources

  1. NPR: “Trump Signs AI Safety Order Seeking Voluntary Review of New Models” (June 2, 2026)
  2. TechTimes: “Trump AI Executive Order Signed: Musk and Zuckerberg Lobbied Review Down to 30 Days” (June 2, 2026)
  3. CyberScoop: “Trump Administration Releases Scaled-Back AI Executive Order” (June 2, 2026)
  4. Council on Foreign Relations: “Assessing Trump’s Executive Order on AI Oversight” (June 2026)
  5. The White House: Fact Sheet: “President Donald J. Trump Promotes Advanced Artificial Intelligence Innovation and Security” (June 2, 2026)
  6. NewsBusters: “Sacks Sets Record Straight: New AI Executive Order Is Not FDA for AI” (June 2, 2026)
  7. TechCrunch: “Trump Signs Narrower AI Executive Order After Industry Objections” (June 2, 2026)
  8. ABC News: “Trump Orders US Government to Cut Ties with Anthropic” (February 2026)
  9. State of Surveillance: “White House $9 Billion for Spy Agencies’ AI”
  10. State of Surveillance: “Trump’s Centralized AI Database on Every American”